CN116434087B - Concrete crack identification method and device based on GOA-SVM cooperative algorithm and unmanned aerial vehicle - Google Patents
Concrete crack identification method and device based on GOA-SVM cooperative algorithm and unmanned aerial vehicle Download PDFInfo
- Publication number
- CN116434087B CN116434087B CN202310354991.2A CN202310354991A CN116434087B CN 116434087 B CN116434087 B CN 116434087B CN 202310354991 A CN202310354991 A CN 202310354991A CN 116434087 B CN116434087 B CN 116434087B
- Authority
- CN
- China
- Prior art keywords
- goa
- svm
- crack
- concrete
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 230000006870 function Effects 0.000 claims abstract description 89
- 238000012549 training Methods 0.000 claims abstract description 59
- 238000004364 calculation method Methods 0.000 claims abstract description 36
- 238000007781 pre-processing Methods 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000005457 optimization Methods 0.000 claims description 34
- 239000011159 matrix material Substances 0.000 claims description 30
- 239000013598 vector Substances 0.000 claims description 27
- 238000010606 normalization Methods 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 16
- 230000009467 reduction Effects 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 12
- 239000000306 component Substances 0.000 claims description 9
- 238000009826 distribution Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 7
- 230000005540 biological transmission Effects 0.000 claims description 5
- 238000002790 cross-validation Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 239000012533 medium component Substances 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 238000003672 processing method Methods 0.000 claims 2
- 230000035939 shock Effects 0.000 claims 1
- 238000012706 support-vector machine Methods 0.000 description 14
- 238000001514 detection method Methods 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 5
- 230000002068 genetic effect Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 239000002245 particle Substances 0.000 description 3
- 241000254032 Acrididae Species 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000005336 cracking Methods 0.000 description 2
- 206010021143 Hypoxia Diseases 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a concrete crack identification method based on a GOA-SVM cooperative algorithm and an unmanned aerial vehicle, which comprises the following steps: s1, collecting concrete crack images, and establishing an original crack image set; s2, preprocessing the concrete crack image; s3, calculating entropy, contrast and energy of the preprocessed image, extracting gray values, establishing a training sample library, and carrying out standardized processing on input data of the training sample library; s4, inputting the standardized training sample into a GOA-SVM algorithm, and learning and training a concrete crack recognition model, so that the model has a crack length and width calculation function, and an optimal GOA-SVM concrete crack recognition model is obtained; s5, embedding a GOA-SVM concrete crack identification model into the unmanned aerial vehicle; s6, shooting the site concrete structure by using the unmanned aerial vehicle, inputting an image shot by the unmanned aerial vehicle into a GOA-SVM concrete crack recognition model, completing recognition of the concrete crack by the GOA-SVM concrete crack recognition model, and calculating the length and width of the crack.
Description
Technical Field
The invention relates to the technical field of concrete crack identification, in particular to a concrete crack identification method and device based on a GOA-SVM cooperative algorithm and an unmanned aerial vehicle.
Background
The width and shape of the cracks of the bridge concrete structure are important parameters for judging the technical condition. Most of the existing concrete cracks are time-consuming and labor-consuming by means of manual identification and drawing, vehicle detection technology and the like, and are high in subjectivity and easy to miss and error. In recent years, crack detection technology based on handheld concrete is developed, but a bridge detection vehicle platform is still required for crack detection of long bridges and high piers, and has no obvious advantage compared with traditional manual direct measurement. In particular, in the plateau region along the railway of the Sichuan and Tibetan, the low air pressure, the oxygen deficiency and the strong ultraviolet rays have great limitation on the labor intensity, and the concrete crack identification and detection method with low labor intensity and simple operation is urgently needed.
In recent years, unmanned planes are widely applied to civil fields such as aviation photographing, geological measurement, high-speed railway management and the like, and a new means is provided for detecting concrete cracks of long, large and high bridges. Through carrying high definition camera, laser platform, damping original paper on unmanned aerial vehicle, can realize the clear shooting of concrete crack with the help of unmanned aerial vehicle to can realize the automatic discernment and the extraction of crack with the help of image processing technique. The technical specification of T/CECS1114-2022 engineering structure digital image method detection standardizes the basic flow and basic operation points of image crack extraction, but fails to give a good crack recognition algorithm, and a training sample comprising what characteristic values is established is not mentioned.
The computer vision measurement technology adopts non-contact and nondestructive detection technology, has the characteristics of high resolution, strong universality, high efficiency and the like, and has wide application prospect in the field of automatic detection of concrete cracks. Currently, thresholding, edge (edge) algorithms, matching (matching) algorithms, C-clustering, fuzzy algorithms, etc. are used to identify concrete cracks. Unlike the above method, machine learning is to learn and automatically extract features through a large number of samples, so as to realize crack identification and extraction, and has better adaptability and generalization prediction capability. The vector holding machine SVM (Support Vector Machine) is a machine learning method based on a statistical learning theory, has a good effect in solving the problems of small samples, high dimensionality, nonlinearity, local minimum points and the like, solves the technical problems of strong dependence on artificial features, weak characterization capability and the like of the traditional machine learning algorithm, such as an artificial neural network, and has a good effect on classification recognition. However, the penalty function number c of the SVM and the width parameter g of the radial basis function have a larger influence on the learning and generalization capability of the SVM model, and are easy to cause under learning or over learning, thereby influencing the recognition effect of concrete cracks. The c and g values determined by traditional cross validation are easy to fall into local optimum, and the objective function pair parameters are required to be tiny; the methods such as particle swarm optimization, genetic algorithm and the like have no outstanding convergence and global optimizing capability when optimizing c and g. Therefore, how to obtain the optimal c and g becomes an important link in realizing concrete crack identification by using the SVM model.
Grasshopper swarm intelligent global algorithm (Grasshopper Optimization Algorithm, GOA) was a new optimization algorithm proposed in 2017. Research shows that compared with random global optimization algorithms such as genetic algorithm, particle swarm, differential evolution algorithm and the like, the GOA algorithm has stronger global optimization capability, faster convergence speed and fewer input parameters. The optimizing process of the GOA algorithm consists of a global optimizing process and a local optimizing process, can effectively reduce the target function calling times of the local optimizing process, and is superior to genetic algorithm, particle swarm algorithm and other algorithms.
Disclosure of Invention
The invention aims to provide a concrete crack identification method based on a GOA-SVM cooperative algorithm and an unmanned aerial vehicle, which can effectively improve the identification efficiency and accuracy of concrete cracks.
In order to achieve the above purpose, the invention provides a concrete crack identification method based on a GOA-SVM cooperative algorithm and an unmanned aerial vehicle, which comprises the following steps:
s1, collecting concrete crack images, and establishing an original crack image set;
s2, preprocessing the concrete crack image;
s3, calculating entropy, contrast and energy of the preprocessed image, extracting gray values, establishing a training sample library, and carrying out standardized processing on input data of the training sample library;
s4, inputting the standardized training sample into a GOA-SVM algorithm, and learning and training a concrete crack recognition model, so that the model has a crack length and width calculation function, and an optimal GOA-SVM concrete crack recognition model is obtained;
s5, embedding a GOA-SVM concrete crack identification model into the unmanned aerial vehicle;
s6, shooting the site concrete structure by using the unmanned aerial vehicle, inputting an image shot by the unmanned aerial vehicle into a GOA-SVM concrete crack recognition model, completing recognition of the concrete crack by the GOA-SVM concrete crack recognition model, and calculating the length and width of the crack.
In a preferred embodiment, in step S1, various concrete crack images are captured by an unmanned aerial vehicle and collected in literature; in the step S2, preprocessing the concrete crack image comprises image enhancement, noise reduction and graying, wherein the image enhancement adopts gray level linear transformation, the noise reduction adopts wavelet transformation, and the graying enables the gray level value of the image to be between 0 and 255.
In a preferred embodiment, in step S3, entropy, contrast and energy of the preprocessed image are calculated, gray values are extracted, a training sample library is established, and input data of the training sample library is normalized, which specifically includes: the entropy, contrast, energy and gray value of gray image are used as the input vector x of training sample i And creates training samples (x i ,y i ) Wherein x is i ＝[x i1 ,x i2 ,x i3 ,x i4 ]The medium components correspond to entropy, contrast, energy and gray value, y i E (o, +) is the identity of the input feature vector, o is non-crack, and +is crack.
In a preferred embodiment, the step S3 specifically includes the following steps:
s31, taking a pixel i as a center in a gray image, taking 3 multiplied by 3 pixels as a calculation window, calculating gray level co-occurrence matrixes P (r, t) in 4 directions, and normalizing to obtain a normalized gray level co-occurrence matrix P (r, t):
P(r,t)＝#{((k 1 ，k 2 ),(l 1 ,l 2 ))∈(L x ×L y )×(L x ×L y )|d,θ,f(k 1 ，k 2 )＝r,f(l 1 ，l 2 ) =t } formula (1)
Wherein Ng is gray level, and the value is 256; p (r, t) is a gray level co-occurrence matrix of Ng×Ng, (L) x ×L y ) For the range domain defined by the gray level co-occurrence matrix, d represents the pixel distance (taking d=1), θ represents the calculation direction (taking θ=0 °, 45 °, 90 °, 135 °), f (k) 1 ，k 2 )＝r，f(l 1 ，l 2 ) T is the gray value of the corresponding rank of the gray image f; # represents the number of pixels established in { }; p (r, t) is a normalized gray level co-occurrence matrix of the gray level co-occurrence matrix P (r, t);
s32, calculating energy values asm, entropy ent and contrast con of the 4-direction normalized gray level co-occurrence matrix p (r, t), and taking the average value of the 4 directions of the obtained values as 3 characteristic values of the window center pixel i:
s33, combining asm, ent, con and gray values to form a training sample input vector x i ＝[x i1 ,x i2 ,x i3 ,x i4 ]And performs normalization processing on each component:
wherein x is i,j X is a group i, ’ j Respectively representing initial input vectors and normalized standard values of j-th dimension input features of an i-th sample; u (u) j Sum sigma j Respectively representing the mean value and standard deviation of the j-th dimension characteristic values of all samples, and inputting a vector x after normalization i The method accords with standard normal distribution in each dimension and is used for learning and training of a subsequent crack recognition model.
In a preferred embodiment, in step S4, the normalized training sample is input into a GOA-SVM algorithm, and a concrete crack recognition model is learned and trained, and has the function of calculating the crack length and width, so as to obtain an optimal GOA-SVM concrete crack recognition model, which includes the following steps:
s41, selecting a penalty parameter c and a kernel parameter g as optimization variables, setting an optimization range, using a k-fold cross validation method, and determining an objective function;
s42, determining population NP and maximum allowable iteration step numberT max Algorithm dimension D, linear reduction function e min 、e max Punishment function c and kernel parameter g optimization range in SVM;
s43, initializing random distribution searching population positions, calculating fitness function values of the current population, and selecting position points with optimal fitness values as directions for guiding next optimization; predicting the optimal position of the next generation population by a position updating formula in the iterative optimization process;
next generation population search locationThe update formula is:
linear reduction parameters of search range of GOA algorithm:
wherein L is the total iteration number, and L is the current iteration number;
s44, comparing the real fitness function value of the predicted optimal individual with the real function fitness value of the current optimal individual position, and if the real fitness function value is superior to the current individual, replacing the current optimal individual with the predicted optimal individual, namely updating the current optimal individual position; if the objective function reaches the objective precision, stopping calculation and outputting inverted parameters; otherwise, go back to step S44, carry on the new round of calculation, iterate until the objective function meets the convergence criterion, reach the goal precision;
s45, substituting the optimized c and g into the SVM, establishing a GOA-SVM concrete crack identification model, and calculating to obtain the crack length B and the width S based on the identified crack, wherein the concrete steps are as follows:
wherein d k Representing the distance between the kth pixel point and the (k+1) th pixel point on the center line of the crack image; x is x k 、y k And x k+1 、y k+1 The coordinates of the kth and k+1 pixel points are respectively set.
In (x) n ，y n )、(x' n ，y' n ) The two points of the far end of the vertical line of the crack edge are coordinates.
In a preferred embodiment, the GOA-SVM concrete crack recognition model implements inner product operation in a high-dimensional feature space by introducing a kernel function, and supposedly maps a sample from the sample space to the high-dimensional space by a certain mapping, and then implements classification in the high-dimensional feature space by adopting a linear method, thereby completing recognition of the concrete crack, and the method comprises the following steps:
establishing a crack identification objective function: max Q (alpha),
in the method, in the process of the invention,
wherein a is i Is the Lagrangian operator corresponding to the ith sample and satisfies a simultaneously i Not less than 0 and Sigma y i a i >0, Σ is a sum function, K (x i ,x j ) As a kernel function, K represents the inner product;
the recognition problem of concrete cracks is converted into GOA-SVM classification problem, and the classification function is as follows:
wherein a is i * Corresponding to one component in the Q (a) optimal solution,sgn is a sign function and returns the result y i E (o, +) when outputting the value y i O, indicating no crack, when the output value y i And +, denotes a crack.
In a preferred embodiment, in step S5, the unmanned aerial vehicle is loaded with a CA103 high-definition camera, and a shock-absorbing connection device is disposed between the high-definition camera and the unmanned aerial vehicle.
The invention also provides a concrete crack identification device based on the GOA-SVM cooperative algorithm and the unmanned aerial vehicle, which comprises the following steps:
the preprocessing module comprises an enhancement unit, a denoising unit and a graying unit, wherein the enhancement unit is used for improving the quality of an image shot by the unmanned aerial vehicle, enriching the information quantity of the image and enhancing the interpretation and identification effects of the image, the denoising unit is used for carrying out noise reduction processing on the image acquired by the unmanned aerial vehicle, and the graying unit is used for converting the gray value of the image shot by the enhanced and denoised unmanned aerial vehicle to be between 0 and 255 so as to form a gray image;
a feature value extraction module including a feature value calculation unit for calculating energy, contrast, gray, and entropy of the gray image, a normalization unit for normalizing the energy, contrast, gray, and entropy, and a sample construction unit for establishing a training sample and a test sample (x i ,y i ) Input vector x of prediction samples i ＝[x i1 ,x i2 ,x i3 ,x i4 ]；
The GOA-SVM optimizing and identifying module is used for training samples and test samples constructed by the characteristic value extracting module and predicting samples x i ＝[x i1 ,x i2 ,x i3 ,x i4 ]The GOA-SVM optimizing and identifying module comprises a GOA optimizing unit, a kernel function setting unit, a GOA-SVM training unit, a GOA-SVM predicting unit, a length calculating unit and a width calculating unit, wherein the GOA optimizing unit is used for setting initial parameters of a GOA algorithm, including population number, maximum allowable iteration step number, algorithm dimension and linear reduction parameter, and the kernel function setting unit is used for setting a kernel function selected by a GOA-SVM concrete crack identifying model; the GOA-SVM training unit is used for training samples (x i ,y i ) Study was performed and test samples (x i ,y i ) Testing; the GOA-SVM prediction unit is used for identifying an input concrete image, and the length calculation unit and the width calculation unit are respectively used for calculating the length and width information of the concrete crack;
the transmission module is used for transmitting data among the crack identification model, the unmanned aerial vehicle, the modules and units in the device; and
and the storage module is used for storing the unmanned aerial vehicle image and the intermediate data for identifying the concrete cracks.
In a preferred embodiment, the feature value calculating unit calculates the energy, contrast, gray level, entropy of the gray level image using formulas (1) to (5), specifically:
taking a pixel i as a center in a gray image, taking 3 multiplied by 3 pixels as a calculation window, calculating gray co-occurrence matrixes P (r, t) in 4 directions, and normalizing to obtain a normalized gray co-occurrence matrix P (r, t):
P(r,t)＝#{((k 1 ，k 2 ),(l 1 ,l 2 ))∈(L x ×L y )×(L x ×L y )|d,θ,f(k 1 ，k 2 )＝r,f(l 1 ，l 2 ) =t } formula (1)
Wherein Ng is gray level, and the value is 256; p (r, t) is a gray level co-occurrence matrix of Ng×Ng, (L) x ×L y ) For the range domain defined by the gray level co-occurrence matrix, d represents the pixel distance (taking d=1), θ represents the calculation direction (taking θ=0 °, 45 °, 90 °, 135 °), f (k) 1 ，k 2 )＝r，f(l 1 ，l 2 ) T is the gray value of the corresponding rank of the gray image f; # represents the number of pixels established in { }; p (r, t) is a normalized gray level co-occurrence matrix of the gray level co-occurrence matrix P (r, t);
calculating energy values asm, entropy ent and contrast con of the 4-direction normalized gray level co-occurrence matrix p (r, t), and taking the average value of the 4 directions of the obtained values as 3 characteristic values of the window center pixel i:
the normalization unit adopts a formula (6) to normalize the energy, contrast, gray level and entropy of the gray level image calculated by the formulas (1) - (5), and specifically comprises the following steps:
combining asm, ent, con and gray values to construct a training sample input vector x i ＝[x i1 ,x i2 ,x i3 ,x i4 ]And performs normalization processing on each component:
wherein x is i,j X is a group i, ’ j Respectively representing initial input vectors and normalized standard values of j-th dimension input features of an i-th sample; u (u) j Sum sigma j Respectively representing the mean value and standard deviation of the j-th dimension characteristic values of all samples, and inputting a vector x after normalization i The method accords with standard normal distribution in each dimension and is used for learning and training of a subsequent crack recognition model.
In a preferred embodiment, the denoising unit is built with denoising methods including wiener filtering, linear filtering, median filtering and wavelet method, and proper denoising method is selected by trial and error method; the kernel function setting unit is provided with radial basis, covariance and linear kernel functions, and adopts a trial-and-error method to select the type of the kernel functions.
Compared with the prior art, the invention has the beneficial effects that: according to the method, a crack training sample library rich enough is built by collecting on-site concrete cracking images in advance, meanwhile, a k-fold cross validation method is utilized to train an SVM image classification model, and GOA algorithm optimization is adopted, so that the problems of optimization and rapid convergence of an SVM algorithm punishment function number c and a radial basis function width parameter g are solved, and compared with a traditional crack segmentation model, the GOA-SVM concrete crack identification model built by the method is stronger in learning ability and better in generalization ability, and the bridge concrete crack real-time detection efficiency is improved; according to the invention, the unmanned aerial vehicle aerial photography technology is combined, the built GOA-SVM concrete crack recognition model is deployed on the unmanned aerial vehicle and used for concrete image recognition, a new shortcut is provided for image recognition, a concrete crack result is obtained quickly, conveniently and efficiently, and quantized output is realized, the labor intensity of concrete crack detection operation is reduced, and a more efficient and accurate detection means is provided for large, long and high concrete structures, particularly for plateau concrete structures.
Drawings
FIG. 1 is a flow chart of concrete crack identification according to a preferred embodiment of the present invention;
FIG. 2 is a graph showing the classification of concrete cracks in a GOA-SVM concrete crack recognition model according to a preferred embodiment of the present invention;
FIG. 3 is a block diagram of a device according to a preferred embodiment of the present invention;
fig. 4 is a unit view of an apparatus according to a preferred embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below. Embodiments of the present invention are intended to be within the scope of the present invention as defined by the appended claims.
Example 1
As shown in fig. 1-2, the embodiment takes actual engineering as an example to describe in detail, and identifies a certain pier concrete near a new bridge of a Tibetan railway, wherein the pier is 30m higher, and local cracking occurs after construction is completed for a period of time.
The concrete crack identification method based on the GOA-SVM cooperative algorithm and the unmanned aerial vehicle comprises the following steps:
and S1, collecting a large number of concrete crack images, and establishing an original image set with better definition.
And S2, carrying out linear enhancement, noise reduction and graying pretreatment on the collected images, wherein the image enhancement adopts gray linear transformation, and the change can enhance the gray contrast of the images and better reserve the edge information. The noise reduction adopts wavelet transform, and the graying makes the gray value of the image between 0 and 255.
Step S3, selecting n typical training samples (n=500) as a sample set, calculating entropy, contrast, energy and gray values, and taking the four values as input vectors x of the training samples i i And constructs a training sample set (x i ,y i ) The method comprises the steps of carrying out a first treatment on the surface of the Wherein x is i ＝[x i1 ,x i2 ,x i3 ,x i4 ]The medium components correspond to entropy, contrast, energy and gray value, y i E (o, +) is the input vector corresponding target value, "o" indicates non-fracture and "+" indicates fracture.
Specifically, the implementation process of step S3 is as follows:
s31, taking a pixel i as a center in a gray image, taking 3 multiplied by 3 pixels as a calculation window, calculating gray level co-occurrence matrixes P (r, t) in 4 directions, and normalizing to obtain a normalized gray level co-occurrence matrix P (r, t):
P(r,t)＝#{((k 1 ，k 2 ),(l 1 ,l 2 ))∈(L x ×L y )×(L x ×L y )|d,θ,f(k 1 ，k 2 )＝r,f(l 1 ，l 2 ) =t } formula (1)
Wherein Ng is gray level, and the value is 256; p (r, t) is a gray level co-occurrence matrix of Ng×Ng, (L) x ×L y ) For the range domain defined by the gray level co-occurrence matrix, d represents the pixel distance (taking d=1), θ represents the calculation direction (taking θ=0 °, 45 °, 90 °, 135 °), f (k) 1 ，k 2 )＝r，f(l 1 ，l 2 ) T is the gray value of the corresponding rank of the gray image f; # represents the number of pixels established in { }; p (r, t) is a normalized gray level co-occurrence matrix of the gray level co-occurrence matrix P (r, t);
s32, calculating energy values asm, entropy ent and contrast con of the 4-direction normalized gray level co-occurrence matrix p (r, t), and taking the average value of the 4 directions of the obtained values as 3 characteristic values of the window center pixel i:
s33, combining asm, ent, con and gray values to form a training sample input vector x i ＝[x i1 ,x i2 ,x i3 ,x i4 ]And performs normalization processing on each component:
wherein x is i,j X's' i,j Respectively representing initial input vectors and normalized standard values of j-th dimension input features of an i-th sample; u (u) j Sum sigma j Respectively representing the mean value and standard deviation of the j-th dimension characteristic values of all samples, and inputting a vector x after normalization i The method accords with standard normal distribution in each dimension and is used for learning and training of a subsequent crack recognition model.
And S4, building a GOA-SVM concrete crack identification model.
Specifically, the implementation process of the GOA-SVM concrete crack identification model in the step S4 is as follows:
step S41, selecting penalty parameters c and kernel parameters g as optimization variables, setting an optimization range, c E (1 e) -2 ,3)，g∈(1e -2 100); the objective function fitness=m/mx100% is chosen as Fitness function. Wherein M is the number of samples counted accurately in recognition, and M is the total number of recognized samples;
step S42, determining the population np=10, the maximum allowable iteration step T max Algorithm dimension d=2, e =100 max ＝1、e min =0.01; penalty function cmax c in SVM max =3, minimum value c min =0.01; nuclear parameter g maximum g max =100, minimum g min ＝0.01；
S43, initializing random distribution searching population positions, calculating fitness function values of the current population, and selecting position points with optimal fitness values as directions for guiding next optimization;
and predicting the optimal position of the next generation population by a position updating formula in the iterative optimization process.
Preferably, the next generation population search location update formula is:
preferably, the correlation coefficient of the search range of the GOA algorithm is determined:
wherein L is the total iteration number, and L is the current iteration number.
And S44, predicting the real fitness function value of the optimal individual and the real function of the current optimal individual position. Comparing the fitness values, if the fitness values are superior to the current individuals, replacing the current optimal individuals by the predicted optimal individuals, namely updating the current optimal individual positions; if the objective function reaches the objective precision, stopping calculation and outputting inverted parameters; otherwise, go back to step S44, perform a new round of calculation, and repeat continuously until the objective function meets the convergence criterion and reaches the objective precision.
And S45, substituting c and g optimized by a GOA algorithm into the SVM, and establishing a GOA-SVM crack identification model. The model can realize automatic crack identification, calculates the crack length B and the crack width S, and specifically comprises the following steps:
wherein d k Representing the distance between the kth pixel point and the (k+1) th pixel point on the center line of the crack image; x is x k 、y k And x k+1 、y k+1 The coordinates of the kth and k+1 pixel points are respectively set.
In (x) n ，y n )、(x' n ，y' n ) The two points of the far end of the vertical line of the crack edge are coordinates.
The GOA-SVM concrete crack identification model realizes inner product operation in a high-dimensional feature space by introducing a kernel function, and presumes that a certain mapping maps a sample from the sample space to the high-dimensional space, and then realizes classification in the high-dimensional feature space by adopting a linear method, thereby completing the identification of the concrete crack, and the GOA-SVM concrete crack identification model comprises the following steps:
establishing a crack identification objective function: max Q (alpha),
in the method, in the process of the invention,
wherein a is i For the Bragg corresponding to the ith sampleThe Langerhans, and simultaneously satisfy a i Not less than 0 and Sigma y i a i >0, Σ is a sum function, K (x i ,x j ) As a kernel function, K represents the inner product.
Further, the recognition problem of the concrete cracks is converted into the classification problem of the GOA-SVM, and the classification function is as follows:
wherein a is i * Corresponding to one component in the Q (a) optimal solution,sgn is a sign function and returns the result y i E (o, +) when outputting the value y i O, indicating no crack, when the output value y i The number +represents a crack.
Preferably, in the invention, the kernel function of the GOA-SVM concrete crack identification model selects a radial basis function REF, and the calculation formula is as follows:
as comparative verification, the present invention compares the currently widely applied support vector machines SVM, GA-SVM and PSO-SVM for crack recognition training and recognition effect, and the results are shown in Table 1. Therefore, the GOA-SVM concrete crack recognition model has more outstanding performance in the aspects of parameter local and global optimization and operation time, and more accurate in accuracy.
Table 1 four algorithm crack recognition algorithms calculate time consuming and recognition accuracy comparisons.
And S5, embedding the GOA-SVM concrete crack identification model into the unmanned aerial vehicle.
And S, planning an unmanned mechanism path, making a flight scheme, shooting a concrete crack of a certain 30m high pier of the Chuancangjingdu bridge, and controlling the general distance between the unmanned plane and the pier to be about 1.5-3 m. And finishing crack identification, length and width information calculation and output through a GOA-SVM crack identification model.
Example 2
As shown in fig. 3 and 4, the invention further provides a concrete crack recognition device based on a GOA-SVM cooperative algorithm and an unmanned aerial vehicle, which comprises: the device comprises a transmission module Z1, a storage module Z2, a preprocessing module Z3, a characteristic value extraction module Z4 and a GOA-SVM optimization and identification module Z5.
The transmission module Z1 is used for mutually transmitting data among the crack identification model, the unmanned aerial vehicle, each module in the device and the units.
The storage module Z2 is used for storing the unmanned aerial vehicle image and the concrete crack identification intermediate data.
The preprocessing module Z3 is used for performing pre-processing on images shot by the unmanned aerial vehicle, enhancing and improving the quality of the images, and graying the images. The preprocessing module Z3 may execute step S2 of the concrete crack recognition method based on the GOA-SVM cooperation algorithm and the unmanned aerial vehicle proposed in embodiment 1 of the present invention. The preprocessing module Z3 comprises an enhancement unit Z3-1, a denoising unit Z3-2 and a Z3-3 graying unit. The enhancement unit Z3-1 is used for improving the quality of images shot by the unmanned aerial vehicle, enriching the information quantity of the images and enhancing the interpretation and recognition effects of the images. The denoising unit Z3-2 is used for denoising images acquired by the unmanned aerial vehicle, such as Gaussian noise and spiced salt noise, so that influence of noise on crack training and recognition accuracy is reduced. The denoising unit Z3-2 is internally provided with common methods such as Wiener filtering, linear filtering, median filtering, wavelet method and the like, and can obtain a good denoising effect by adopting a trial and error method. The graying unit Z3-3 is used for converting the gray value of the image shot by the enhanced and denoised unmanned aerial vehicle to be between 0 and 255, so that the image is a gray image.
The feature value extraction module Z4 is used for extracting features of the imageThe feature value, establish the input vector of training sample and predictive sample, the feature value draws the module Z4 to include: a characteristic value calculation unit Z4-1, a normalization unit Z4-2 and a sample construction unit Z4-3. The feature value calculating unit Z4-1 is configured to calculate energy, contrast, gray level, and entropy of the gray level image, and specifically adopts formulas (1) to (5). The normalization unit Z4-2 is used to normalize energy, contrast, gray scale, entropy, specifically using equation (6). A sample construction unit Z4-3 for constructing training samples (x i ,y i ) Input vector x of prediction samples i ＝[x i1 ,x i2 ,x i3 ,x i4 ]。
GOA-SVM optimization and recognition module Z5 is used to train and test samples (x) constructed by training and test feature value extraction module Z4 i ,y i ) And for prediction sample x i ＝[x i1 ,x i2 ,x i3 ,x i4 ]And performing prediction classification. The module can execute the step S4 of the concrete crack identification method based on the GOA-SVM cooperative algorithm and the unmanned aerial vehicle, which is provided by the embodiment 1 of the invention. The GOA-SVM optimization and identification module Z5 comprises: a GOA optimizing unit Z5-1, a kernel function setting unit Z5-2, a GOA-SVM training unit Z5-3, a GOA-SVM predicting unit Z5-4, a length calculating unit Z5-5 and a width calculating unit Z5-6.
The GOA optimization unit Z5-1 is configured to set initial parameters of a GOA algorithm, including a population np=10, a maximum allowable iteration step T max The algorithm dimension D, including the adaptive parameter maximum c =100 max =3, adaptive parameter min c min =0.01, etc., the SVM model hyper-parameters c, g are optimized. The kernel function setting unit Z5-2 is used for setting the kernel function selected by the GOA-SVM. The unit is provided with radial basis, covariance, linear kernel functions and the like, and a user can select the kernel function type by adopting a trial-and-error method to obtain a good mapping effect. GOA-SVM training unit Z5-3 is used to train the pre-processed training samples (x i ,y i ). In the training process, 90% of the data in the preprocessed training samples in the sample library are randomly used as a training data set, and the other 10% of the data are used as a test data set. The GOA-SVM prediction unit Z5-4 is used to recognize an inputted concrete image. Length calculation sheetThe element Z5-5 and the width calculation unit Z5-6 are respectively used for calculating the length and width information of the concrete crack.
Preferably, the interactive links among the modules are fed back through the transmission module Z1. Thus, to ensure link stability, wireless, wired, or hybrid conduction may be implemented depending on field condition requirements.
The concrete implementation process shows that the concrete crack identification method and device based on the GOA-SVM cooperative algorithm and the unmanned aerial vehicle provided by the invention have the advantages of simplicity, high efficiency, high prediction precision and the like, and can effectively solve the problems of high labor intensity and poor precision of the current concrete crack identification.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (5)
1. A concrete crack identification method based on a GOA-SVM cooperative algorithm and an unmanned aerial vehicle is characterized by comprising the following steps of: the method comprises the following steps:
s1, collecting concrete crack images, and establishing an original crack image set;
s2, preprocessing the concrete crack image;
s3, calculating entropy, contrast and energy of the preprocessed image, extracting gray values, establishing a training sample library, and carrying out standardized processing on input data of the training sample library;
s4, inputting the standardized training sample into a GOA-SVM algorithm, and learning and training a concrete crack recognition model, so that the model has a crack length and width calculation function, and an optimal GOA-SVM concrete crack recognition model is obtained;
s5, embedding a GOA-SVM concrete crack identification model into the unmanned aerial vehicle;
s6, shooting the site concrete structure by using an unmanned aerial vehicle, inputting an image shot by the unmanned aerial vehicle into a GOA-SVM concrete crack recognition model, completing recognition of the concrete crack by the GOA-SVM concrete crack recognition model, and calculating the length and width of the crack;
in step S3, calculating entropy, contrast and energy of the preprocessed image, extracting gray values, establishing a training sample library, and performing standardization processing on input data of the training sample library, which specifically includes: the entropy, contrast, energy and gray value of gray image are used as the input vector x of training sample i And creates training samples (x i ,y i ) Wherein x is i ＝[x i1 ,x i2 ,x i3 ,x i4 ]The medium components correspond to entropy, contrast, energy and gray value, y i E (o, +) is the identification of the input feature vector, o is non-crack, and +is crack;
step S3, specifically comprising the following steps:
s31, taking a pixel i in a gray image as a center, taking 3 multiplied by 3 pixels as a calculation window, calculating gray level co-occurrence matrixes P (r, t) in 4 directions, and normalizing to obtain a normalized gray level co-occurrence matrix P (r, t):
P(r,t)＝#{((k 1 ，k 2 ),(l 1 ,l 2 ))∈(L x ×L y )×(L x ×L y )|d,θ,f(k 1 ，k 2 )
＝r,f(l 1 ，l 2 ) =t } formula (1)
Wherein Ng is gray level, and the value is 256; p (r, t) is a gray level co-occurrence matrix of Ng×Ng, (L) x ×L y ) The range domain defined for the gray co-occurrence matrix, d represents the pixel distance, θ represents the calculation direction, f (k) 1 ，k 2 )＝r，f(l 1 ，l 2 ) T is the gray value of the corresponding rank of the gray image f; # represents the number of pixels established in { }; p (r, t) is a normalized gray level co-occurrence matrix of the gray level co-occurrence matrix P (r, t);
s32, calculating energy values asm, entropy ent and contrast con of the 4-direction normalized gray level co-occurrence matrix p (r, t), and taking the average value of the 4 directions of the obtained values as 3 characteristic values of the window center pixel i:
s33, combining asm, ent, con and gray values to form a training sample input vector x i ＝[x i1 ,x i2 ,x i3 ,x i4 ]And performs normalization processing on each component:
wherein x is i,j X is a group i, ’ j Respectively representing initial input vectors and normalized standard values of j-th dimension input features of an i-th sample; u (u) j Sum sigma j Respectively representing the mean value and standard deviation of the j-th dimension characteristic values of all samples, and inputting a vector x after normalization i The method accords with standard normal distribution in each dimension and is used for learning and training a subsequent crack recognition model;
in step S4, inputting the standardized training sample into a GOA-SVM algorithm, learning and training a concrete crack identification model, and enabling the model to have the crack length and width calculation function, so as to obtain an optimal GOA-SVM concrete crack identification model, wherein the method comprises the following steps of:
s41, selecting a penalty parameter c and a kernel parameter g as optimization variables, setting an optimization range, using a k-fold cross validation method, and determining an objective function;
s42, determining population NP and maximum allowable iteration step number T max Algorithm dimension D, linear reduction function e min 、e max Punishment function c and kernel parameter g optimization range in SVM;
s43, initializing random distribution searching population positions, calculating fitness function values of the current population, and selecting position points with optimal fitness values as directions for guiding next optimization; predicting the optimal position of the next generation population by a position updating formula in the iterative optimization process;
next generation population search locationThe update formula is:
linear reduction parameters of search range of GOA algorithm:
wherein L is the total iteration number, and L is the current iteration number;
s44, comparing the real fitness function value of the predicted optimal individual with the real function fitness value of the current optimal individual position, and if the real fitness function value is superior to the current individual, replacing the current optimal individual with the predicted optimal individual, namely updating the current optimal individual position; if the objective function reaches the objective precision, stopping calculation and outputting inverted parameters; otherwise, go back to step S44, carry on the new round of calculation, iterate until the objective function meets the convergence criterion, reach the goal precision;
s45, substituting the optimized c and g into the SVM, establishing a GOA-SVM concrete crack identification model, and calculating to obtain the crack length B and the width S based on the identified crack, wherein the concrete steps are as follows:
wherein d k Representing the distance between the kth pixel point and the (k+1) th pixel point on the center line of the crack image; x is x k 、y k And x k+1 、y k+1 The coordinates of the kth pixel point and the k+1 pixel point are respectively;
in (x) n ，y n )、(x’ n ，y’ n ) The two-point coordinates of the farthest end of the vertical line of the crack edge;
the GOA-SVM concrete crack identification model realizes inner product operation in a high-dimensional feature space by introducing a kernel function, and a certain mapping is assumed to map a sample from the sample space to the high-dimensional space, and then classification is realized in the high-dimensional feature space by adopting a linear method, so that the identification of the concrete crack is completed, and the GOA-SVM concrete crack identification model comprises the following steps:
establishing a crack identification objective function: max Q (a),
in the method, in the process of the invention,
wherein a is i Is the Lagrangian operator corresponding to the ith sample and satisfies a simultaneously i Not less than 0 and Sigma y i a i >0, Σ is a sum function, K (x i ,x j ) As a kernel function, K represents the inner product;
the recognition problem of concrete cracks is converted into GOA-SVM classification problem, and the classification function is as follows:
wherein a is i * Corresponding to one component in the Q (a) optimal solution,sgn is a sign function and returns the result y i E (o, +); when the output value y i When o is zero, the non-crack is represented, when the value y is output i When +is, the crack is indicated.
2. The concrete crack identification method based on the GOA-SVM cooperative algorithm and the unmanned aerial vehicle, which is disclosed in claim 1, is characterized in that: in the step S1, various concrete crack images are shot through an unmanned aerial vehicle and collected in literature; in the step S2, preprocessing the concrete crack image comprises image enhancement, noise reduction and graying, wherein the image enhancement adopts gray level linear transformation, the noise reduction adopts wavelet transformation, and the graying enables the gray level value of the image to be between 0 and 255.
3. The concrete crack identification method based on the GOA-SVM cooperative algorithm and the unmanned aerial vehicle, which is characterized in that: in step S5, the unmanned aerial vehicle that adopts is loaded with the CA103 high definition camera, sets up shock attenuation connecting device between high definition camera and the unmanned aerial vehicle.
4. A concrete crack recognition device based on GOA-SVM cooperation algorithm and unmanned aerial vehicle, its characterized in that: comprising the following steps:
the preprocessing module comprises an enhancement unit, a denoising unit and a graying unit, wherein the enhancement unit is used for improving the quality of an image shot by the unmanned aerial vehicle, enriching the information quantity of the image and enhancing the interpretation and identification effects of the image, the denoising unit is used for carrying out noise reduction processing on the image acquired by the unmanned aerial vehicle, and the graying unit is used for converting the gray value of the image shot by the enhanced and denoised unmanned aerial vehicle to be between 0 and 255 so as to form a gray image;
the characteristic value extraction module comprises a characteristic value calculation unit, a normalization unit and a sample construction unit, wherein the characteristic value calculation unit is used for calculating energy, contrast, gray level and entropy of a gray level image, the normalization unit is used for normalizing the energy, the contrast, the gray level and the entropy, and the sample construction unit is used for establishing a gray level imageTraining samples and test samples (x i ,y i ) Input vector x of prediction samples i ＝[x i1 ,x i2 ,x i3 ,x i4 ]Wherein x is i ＝[x i1 ,x i2 ,x i3 ,x i4 ]The medium component corresponds to entropy, contrast, energy and gray value respectively;
the GOA-SVM optimizing and identifying module is used for training and testing the training samples and the testing samples constructed by the characteristic value extracting module and inputting vectors x of the predicted samples i ＝[x i1 ,x i2 ,x i3 ,x i4 ]Performing prediction classification, wherein the GOA-SVM optimization recognition module comprises a GOA optimization unit, a kernel function setting unit, a GOA-SVM training unit, a GOA-SVM prediction unit, a length calculation unit and a width calculation unit, wherein the GOA optimization unit is used for setting initial parameters of a GOA algorithm, including population number, maximum allowable iteration step number, algorithm dimension and linear reduction parameter, and the kernel function setting unit is used for setting a kernel function selected by a GOA-SVM concrete crack recognition model; the GOA-SVM training unit is used for training samples (x i ,y i ) Study was performed and test samples (x i ,y i ) Testing; the GOA-SVM prediction unit is used for identifying an input concrete image, and the length calculation unit and the width calculation unit are respectively used for calculating the length information and the width information of the concrete crack;
the transmission module is used for transmitting data among the crack identification model, the unmanned aerial vehicle, the modules and units in the device; and
the storage module is used for storing the unmanned aerial vehicle image and the intermediate data for identifying the concrete cracks;
the characteristic value calculating unit calculates the energy, contrast, gray level and entropy of the gray level image by adopting formulas (1) to (5), and specifically comprises the following steps:
taking a pixel i as a center in a gray image, taking 3 multiplied by 3 pixels as a calculation window, calculating gray co-occurrence matrixes P (r, t) in 4 directions, and normalizing to obtain a normalized gray co-occurrence matrix P (r, t):
P(r,t)＝#{((k 1 ，k 2 ),(l 1 ,l 2 ))∈(L x ×L y )×(L x ×L y )|d,θ,f(k 1 ，k 2 )
＝r,f(l 1 ，l 2 ) =t } formula (1)
Wherein Ng is gray level, and the value is 256; p (r, t) is a gray level co-occurrence matrix of Ng×Ng, (L) x ×L y ) The range domain defined for the gray co-occurrence matrix, d represents the pixel distance, θ represents the calculation direction, f (k) 1 ，k 2 )＝r，f(l 1 ，l 2 ) T is the gray value of the corresponding rank of the gray image f; # represents the number of pixels established in { }; p (r, t) is a normalized gray level co-occurrence matrix of the gray level co-occurrence matrix P (r, t);
calculating energy values asm, entropy ent and contrast con of the 4-direction normalized gray level co-occurrence matrix p (r, t), and taking the average value of the 4 directions of the obtained values as 3 characteristic values of the window center pixel i:
the normalization unit adopts a formula (6) to normalize the energy, contrast, gray level and entropy of the gray level image calculated by the formulas (1) - (5), and specifically comprises the following steps:
combining asm, ent, con and gray values to construct a training sample input vector x i ＝[x i1 ,x i2 ,x i3 ,x i4 ]And to thereinThe components are subjected to standardization processing:
wherein x is i,j X is a group i, ’ j Respectively representing initial input vectors and normalized standard values of j-th dimension input features of an i-th sample; u (u) j Sum sigma j Respectively representing the mean value and standard deviation of the j-th dimension characteristic values of all samples, and inputting a vector x after normalization i The method accords with standard normal distribution in each dimension and is used for learning and training a subsequent crack recognition model;
the construction of the GOA-SVM concrete crack identification model comprises the following steps:
s41, selecting a penalty parameter c and a kernel parameter g as optimization variables, setting an optimization range, using a k-fold cross validation method, and determining an objective function;
s42, determining population NP and maximum allowable iteration step number T max Algorithm dimension D, linear reduction function e min 、e max Punishment function c and kernel parameter g optimization range in SVM;
s43, initializing random distribution searching population positions, calculating fitness function values of the current population, and selecting position points with optimal fitness values as directions for guiding next optimization; predicting the optimal position of the next generation population by a position updating formula in the iterative optimization process;
next generation population search location X i d The update formula is:
linear reduction parameters of search range of GOA algorithm:
wherein L is the total iteration number, and L is the current iteration number;
s44, comparing the real fitness function value of the predicted optimal individual with the real function fitness value of the current optimal individual position, and if the real fitness function value is superior to the current individual, replacing the current optimal individual with the predicted optimal individual, namely updating the current optimal individual position; if the objective function reaches the objective precision, stopping calculation and outputting inverted parameters; otherwise, go back to step S44, carry on the new round of calculation, iterate until the objective function meets the convergence criterion, reach the goal precision;
s45, substituting the optimized c and g into the SVM, establishing a GOA-SVM concrete crack identification model, and calculating to obtain the crack length B and the width S based on the identified crack, wherein the concrete steps are as follows:
wherein d k Representing the distance between the kth pixel point and the (k+1) th pixel point on the center line of the crack image; x is x k 、y k And x k+1 、y k+1 The coordinates of the kth pixel point and the k+1 pixel point are respectively;
in (x) n ，y n )、(x’ n ，y’ n ) The two-point coordinates of the farthest end of the vertical line of the crack edge;
the GOA-SVM concrete crack identification model realizes inner product operation in a high-dimensional feature space by introducing a kernel function, and a certain mapping is assumed to map a sample from the sample space to the high-dimensional space, and then classification is realized by adopting a linear method in the high-dimensional feature space, so that the identification of the concrete crack is completed, and the method comprises the following steps:
establishing a crack identification objective function: max Q (a),
in the method, in the process of the invention,
wherein a is i Is the Lagrangian operator corresponding to the ith sample and satisfies a simultaneously i Not less than 0 and Sigma y i a i >0, Σ is a sum function, K (x i ,x j ) As a kernel function, K represents the inner product;
the recognition problem of concrete cracks is converted into GOA-SVM classification problem, and the classification function is as follows:
wherein a is i * Corresponding to one component in the Q (a) optimal solution,sgn is a sign function and returns the result y i E (o, +); when the output value y i When o is zero, the non-crack is represented, when the value y is output i When +is, the crack is indicated.
5. The concrete crack recognition device based on the GOA-SVM cooperative algorithm and the unmanned aerial vehicle as claimed in claim 4, wherein: the denoising unit is internally provided with a denoising processing method comprising wiener filtering, linear filtering, median filtering and a wavelet method, and a proper denoising processing method is selected by a trial-and-error method; the kernel function setting unit is provided with a radial basis, covariance and a linear kernel function, and adopts a trial-and-error method to select the type of the kernel function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310354991.2A CN116434087B (en) | 2023-04-04 | 2023-04-04 | Concrete crack identification method and device based on GOA-SVM cooperative algorithm and unmanned aerial vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310354991.2A CN116434087B (en) | 2023-04-04 | 2023-04-04 | Concrete crack identification method and device based on GOA-SVM cooperative algorithm and unmanned aerial vehicle |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116434087A CN116434087A (en) | 2023-07-14 |
| CN116434087B true CN116434087B (en) | 2024-01-23 |
Family
ID=87092058
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310354991.2A Active CN116434087B (en) | 2023-04-04 | 2023-04-04 | Concrete crack identification method and device based on GOA-SVM cooperative algorithm and unmanned aerial vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116434087B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108051450A (en) * | 2017-11-01 | 2018-05-18 | 中铁大桥科学研究院有限公司 | A kind of bridge health monitoring system and method based on unmanned plane |
| CN108564569A (en) * | 2018-03-23 | 2018-09-21 | 石家庄铁道大学 | A kind of distress in concrete detection method and device based on multinuclear classification learning |
| CN108985363A (en) * | 2018-07-03 | 2018-12-11 | 长安大学 | A kind of cracks in reinforced concrete bridge classifying identification method based on RBPNN |
| CN110348075A (en) * | 2019-06-20 | 2019-10-18 | 湖南科技大学 | A kind of grinding surface roughness prediction technique based on improvement algorithm of support vector machine |
| CN114937016A (en) * | 2022-05-25 | 2022-08-23 | 广西大学 | Bridge concrete crack real-time detection method and device based on edge calculation and Transformer |
| CN115482462A (en) * | 2022-08-31 | 2022-12-16 | 东南大学 | Concrete crack identification system and method based on deep learning |
| CN115511836A (en) * | 2022-09-28 | 2022-12-23 | 丽水市市政设施管理中心(丽水市节约用水管理中心) | Bridge crack grade evaluation method and system based on reinforcement learning algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018165753A1 (en) * | 2017-03-14 | 2018-09-20 | University Of Manitoba | Structure defect detection using machine learning algorithms |
-
2023
- 2023-04-04 CN CN202310354991.2A patent/CN116434087B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108051450A (en) * | 2017-11-01 | 2018-05-18 | 中铁大桥科学研究院有限公司 | A kind of bridge health monitoring system and method based on unmanned plane |
| CN108564569A (en) * | 2018-03-23 | 2018-09-21 | 石家庄铁道大学 | A kind of distress in concrete detection method and device based on multinuclear classification learning |
| CN108985363A (en) * | 2018-07-03 | 2018-12-11 | 长安大学 | A kind of cracks in reinforced concrete bridge classifying identification method based on RBPNN |
| CN110348075A (en) * | 2019-06-20 | 2019-10-18 | 湖南科技大学 | A kind of grinding surface roughness prediction technique based on improvement algorithm of support vector machine |
| CN114937016A (en) * | 2022-05-25 | 2022-08-23 | 广西大学 | Bridge concrete crack real-time detection method and device based on edge calculation and Transformer |
| CN115482462A (en) * | 2022-08-31 | 2022-12-16 | 东南大学 | Concrete crack identification system and method based on deep learning |
| CN115511836A (en) * | 2022-09-28 | 2022-12-23 | 丽水市市政设施管理中心(丽水市节约用水管理中心) | Bridge crack grade evaluation method and system based on reinforcement learning algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 信息科技.2021,第2021卷(第03期),C034-119. * |
| 张亮.桥梁病害图像处理及识别研究.工程科技Ⅱ辑 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116434087A (en) | 2023-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110569837B (en) | Method and device for optimizing damage detection result | |
| CN111461134B (en) | Low-resolution license plate recognition method based on generation countermeasure network | |
| CN111931637A (en) | Cross-modal pedestrian re-identification method and system based on double-current convolutional neural network | |
| CN112464911A (en) | Improved YOLOv 3-tiny-based traffic sign detection and identification method | |
| CN111860106B (en) | Unsupervised bridge crack identification method | |
| CN110991257B (en) | Polarized SAR oil spill detection method based on feature fusion and SVM | |
| CN113822880A (en) | Crack identification method based on deep learning | |
| CN108345866B (en) | Pedestrian re-identification method based on deep feature learning | |
| CN111223087B (en) | Automatic bridge crack detection method based on generation countermeasure network | |
| CN111126155B (en) | Pedestrian re-identification method for generating countermeasure network based on semantic constraint | |
| CN114283325A (en) | Underwater target identification method based on knowledge distillation | |
| CN114283285A (en) | Cross consistency self-training remote sensing image semantic segmentation network training method and device | |
| CN116796810A (en) | Deep neural network model compression method and device based on knowledge distillation | |
| CN116434087B (en) | Concrete crack identification method and device based on GOA-SVM cooperative algorithm and unmanned aerial vehicle | |
| CN117593243A (en) | Compressor appearance self-adaptive detection method guided by reliable pseudo tag | |
| CN117152094A (en) | Method, device and system for analyzing surface defects of steel plate based on computer vision | |
| CN116524341A (en) | Underwater target identification method based on quantitative distillation | |
| CN115240070A (en) | Crack detection method | |
| CN115310559A (en) | Target identification method based on Bayes-Transformer neural network model | |
| CN114821098A (en) | High-speed pavement damage detection algorithm based on gray gradient fusion characteristics and CNN | |
| CN113313678A (en) | Automatic sperm morphology analysis method based on multi-scale feature fusion | |
| CN107563452B (en) | Cross-domain foundation cloud picture classification method based on discriminant measure learning | |
| CN111652048A (en) | A deep learning based 1: n face comparison method | |
| CN110502992A (en) | A kind of fast face recognition method of the fixed scene video based on relation map | |
| CN112816408B (en) | Flaw detection method for optical lens |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |